Get the newsletter
Nvidia · Software · Filed Nov 17, 2025 · USPTO published Nov 19, 2026 · This week's release

A New Nvidia Patent Speeds Up Game Lighting by Focusing on the Brightest Spots

Nvidia's filing turns a picture into a bumpy 3D surface where the brightest pixels are the biggest targets, then shoots virtual light at it. The sun gets hit most often, which is exactly where a game's lighting needs to look.

A game view with a sun and lake paired with a grid helps graphics chips focus lighting work on the brightest spots. Drawing from patent filing US 2026/0368993 A1.
A game view with a sun and lake paired with a grid helps graphics chips focus lighting work on the brightest spots.All 18 drawings ↓
Publication number
US 2026/0368993 A1(US20260368993A1)
Applicant
Nvidia Corporation
Filed · Published
Nov 17, 2025 · Nov 19, 2026
Inventors
Nathan Morrical
Status when we published
Waiting for an examiner
Document
1 claims
US classification
345/426
Parent application
Continuation of 18/490,171 (filed Oct 19, 2023)(18490171)

What Nvidia's brightness-hill lighting trick does

Every time a game draws a new frame, which can happen 60 or 90 times a second, your graphics chip has to work out how the scene is lit. A sunny sky is the classic headache. The sun supplies most of the light but covers only a few pixels, so checking pixels at random wastes effort on dark ground and dull sky.

Nvidia's patent turns the picture into a bumpy 3D surface, with each pixel's height set by how much it matters for lighting. Then it fires virtual rays of light at that surface. Tall, steep spots get hit far more often, so the sun gets checked constantly and the shadows barely at all.

In the filing's own example, the sun pixel is about six times more likely to be hit than any other pixel. The result is less grainy lighting for the same amount of work.

From the filing · CLAIM 1
… generate, based on the importance values, a three-dimensional (3D) geometric mesh of the image corresponding to interaction probabilities for the one or more pixels, respective heights of cells forming the 3D geometric mesh corresponding to respective interaction probabilities …

Translation: It builds a 3D surface where the height of each cell represents how likely light is to interact with that pixel.

How rays pick pixels off a probability mesh

Lighting a scene realistically means estimating how much light reaches each surface. The usual shortcut is importance sampling (checking the most influential pixels more often than the dull ones). The classic way to do it, inverse transform sampling, turns each row and column of pixel brightness into a running-total curve (a cumulative distribution function) and then uses a binary search (repeatedly halving a list to find a match). The filing says that search behaves unpredictably and is costly at 60 to 90 frames a second.

Claim 1 swaps the search for geometry:

  • Score each pixel of an image for importance.
  • Build a 3D mesh where each cell's height matches that pixel's probability of being picked.
  • Fire simulated light rays at the mesh.
  • Read off which pixel locations the rays hit.

In the description, the running-total curves become ribbons of triangles, so steeper stretches of a curve offer more surface for a ray to hit. The slope where a ray lands also gives the probability needed to keep the final lighting math fair. Because the mesh is made of triangles, dedicated ray tracing hardware (the filing names Nvidia's RT cores) can search it quickly.

The filing also describes optional tweaks: keeping only the most influential control points to simplify the mesh, rescaling it to a 0 to 1 range (45 to 50% faster traversal in its tests), and merging similar neighboring rows (20 to 30% faster).

From the filing · THE ABSTRACT
A geometric mesh can be generated based on a set of cumulative distribution functions (CDFs) for various rows and columns of pixels in a texture, and individual rays can be traced against this mesh, with those rays having a higher probability of intersection at a point with greater importance, such as a higher brightness value.

Translation: Virtual light rays are bounced against this mesh so they naturally hit the brightest and most important spots first.

Why this matters for real-time game lighting

For anyone who plays games on a PC, console or streaming service, the payoff would be lighting that looks cleaner at the same frame rate, or the same look at a higher one. Grainy, flickering light comes from a renderer taking too few useful samples, and this approach makes each sample count for more.

The filing says its methods are light enough to run on a client device such as a gaming console, and the same math can also run on cloud servers that stream the finished picture to you. It also describes the idea as usable beyond games, for other image work. The reported gain is up to 70% faster rendering in one example, with noise reduction similar to the traditional method.

This is the 46th Nvidia filing we've tracked in the GPU rendering race since July, adding to work like shadows and reflections chip and memory shortcut for AI images.

Source: US 2026/0368993 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Nvidia patent news breaks.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings · 18 sheets

18 drawing sheets from US 2026/0368993 A1 · click any drawing to enlarge

Dividing a game scene into lights and pixel grids, drawing sheet 1 from US 2026/0368993 A1
Sheet 1 Dividing a game scene into lights and pixel grids
Pinpointing where the brightest lighting comes from along a row, drawing sheet 2 from US 2026/0368993 A1
Sheet 2 Pinpointing where the brightest lighting comes from along a row
Mapping out light levels across an entire surface grid, drawing sheet 3 from US 2026/0368993 A1
Sheet 3 Mapping out light levels across an entire surface grid
Turning light maps into 3D shapes for faster ray tracing, drawing sheet 4 from US 2026/0368993 A1
Sheet 4 Turning light maps into 3D shapes for faster ray tracing
How the graphics chip turns game instructions into displayed frames, drawing sheet 5 from US 2026/0368993 A1
Sheet 5 How the graphics chip turns game instructions into displayed frames
Steps for lighting objects faster by targeting high brightness areas, drawing sheet 6 from US 2026/0368993 A1
Sheet 6 Steps for lighting objects faster by targeting high brightness areas
Streaming rendered game visuals between cloud servers and player devices, drawing sheet 7 from US 2026/0368993 A1
Sheet 7 Streaming rendered game visuals between cloud servers and player devices
Chip components that calculate and store data for AI models, drawing sheet 8 from US 2026/0368993 A1
Sheet 8 Chip components that calculate and store data for AI models
Show the other 10 sheets
Hardware units paired with memory to calculate lighting faster, drawing sheet 9 from US 2026/0368993 A1
Sheet 9 Hardware units paired with memory to calculate lighting faster
Data center servers running lighting software across many computers, drawing sheet 10 from US 2026/0368993 A1
Sheet 10 Data center servers running lighting software across many computers
Computer chips that route scene data to graphics cards, drawing sheet 11 from US 2026/0368993 A1
Sheet 11 Computer chips that route scene data to graphics cards
A mobile device layout connecting sensors and graphics hardware, drawing sheet 12 from US 2026/0368993 A1
Sheet 12 A mobile device layout connecting sensors and graphics hardware
Inside a computer platform built to render rich 3D scenes, drawing sheet 13 from US 2026/0368993 A1
Sheet 13 Inside a computer platform built to render rich 3D scenes
Multi-core chip layout with integrated graphics and fast shared memory, drawing sheet 14 from US 2026/0368993 A1
Sheet 14 Multi-core chip layout with integrated graphics and fast shared memory
Training an AI model on medical images to automate analysis, drawing sheet 15 from US 2026/0368993 A1
Sheet 15 Training an AI model on medical images to automate analysis
Connecting cloud servers and graphics cards to serve AI models, drawing sheet 16 from US 2026/0368993 A1
Sheet 16 Connecting cloud servers and graphics cards to serve AI models
Retraining an existing AI with new customer images, drawing sheet 17 from US 2026/0368993 A1
Sheet 17 Retraining an existing AI with new customer images
Using AI to label raw images for training, drawing sheet 18 from US 2026/0368993 A1
Sheet 18 Using AI to label raw images for training
Patent filing page

Watchlists this filing belongs to

All watchlists →
Reader comments

No comments yet

Start the discussion

Real name or a handle, either is fine. Comments are read by a person before they appear, so allow a little time. Keep it about the filing.